Scrape and publish Telemetry probe data from Firefox
Перейти к файлу
data-updater 2413db9e99 Update repositories.yaml with new FOG metrics_yamls list 2024-11-18 11:42:24 +01:00
.circleci
.github CI: make it possible to dispatch the fog updater manually 2024-09-12 18:51:13 +02:00
docs
fog-updater fog-updater: Find indentation by looking for the first dash 2024-09-12 18:00:59 +02:00
notebooks
probe_scraper Send email for ping retention errors instead of raising exception (#833) 2024-11-14 19:53:24 +00:00
tests Send email for ping retention errors instead of raising exception (#833) 2024-11-14 19:53:24 +00:00
.dockerignore
.flake8
.gce_boto
.gcloudignore
.gitignore
.yamllint
CODE_OF_CONDUCT.md
CONTRIBUTING.md
Dockerfile
LICENSE
Makefile
README.md
conftest.py
docker-compose.yml
docs.png
main.py
probeinfo_api.yaml
pytest.ini
repositories.yaml Update repositories.yaml with new FOG metrics_yamls list 2024-11-18 11:42:24 +01:00
requirements.txt Update glean-parser requirement from ~=15.1.0 to ~=15.2.0 2024-11-04 16:40:14 +01:00
setup.py
test_requirements.txt

README.md

probe-scraper

Scrape Telemetry probe data from Firefox repositories.

This extracts per-version Telemetry probe data for Firefox and other Mozilla products from registry files like Histograms.json and Scalars.yaml. The data allows answering questions like "which Firefox versions is this Telemetry probe in anyway?". Also, probes outside of Histograms.json - like the CSS use counters - are included in the output data.

The data is pulled from two different sources:

Probe Scraper outputs JSON to https://probeinfo.telemetry.mozilla.org. Effectively, this creates a REST API which can be used by downstream tools like mozilla-schema-generator and various data dictionary type applications (see below).

An OpenAPI reference to this API is available:

probeinfo API docs

A web tool to explore the Firefox-related data is available at probes.telemetry.mozilla.org. A project to develop a similar view for Glean-based data is under development in the Glean Dictionary.

Deprecation

Deprecation is an important step in an application lifecycle. Because of the backwards-compatible nature of our pipeline, we do not remove Glean apps or variants from the repositories.yaml file - instead, we mark them as deprecated.

Marking an App Variant as deprecated

When an app variant is marked as deprecated (see this example from Fenix), the following happens:

Marking an App as deprecated

When an app is marked as deprecated (see this example of Firefox for Fire TV), the following happens:

  • It no longer shows by default in the Glean Dictionary. (Deprecated apps can be viewed by clicking the Show deprecated applications checkbox)

Adding a New Glean Repository

To scrape a git repository for probe definitions, an entry needs to be added in repositories.yaml. The exact format of the entry depends on whether you are adding an application or a library. See below for details.

Adding an application

For a given application, Glean metrics are emitted by the application itself, any libraries it uses that also use Glean, as well as the Glean library proper. Therefore, probe scraper needs a way to find all of the dependencies to determine all of the metrics emitted by that application.

Therefore, each application should specify a dependencies parameter, which is a list of Glean-using libraries used by the application. Each entry should be a library name as specified by the library's library_names parameter.

For Android applications, if you're not sure what the dependencies of the application are, you can run the following command at the root of the project folder:

$ ./gradlew :app:dependencies

See the full application schema documentation for descriptions of all the available parameters.

Adding a library

Probe scraper also needs a way to map dependencies back to an entry in the repositories.yaml file. Therefore, any libraries defined should also include their build-system-specific library names in the library_names parameter.

See the full library schema documentation for descriptions of all the available parameters.

Developing the probe-scraper

You can choose to develop using the container, or locally. Using the container will be slower, since changes will trigger a rebuild of the container. But using the container method will ensure that your PR passes CircleCI build/test phases.

Local development

You may wish to, instead of installing all these requirements in your global Python environment, start by generating and activating a Python virtual environment. The .gitignore expects it to be called ENV or venv:

python -m venv venv
. venv/bin/activate

Install the requirements:

pip install -r requirements.txt
pip install -r test_requirements.txt
python setup.py develop

Run tests. This by default does not run tests that require a web connection:

pytest tests/

To run all tests, including those that require a web connection:

pytest tests/ --run-web-tests

To test whether the code conforms to the style rules, you can run:

python -m black --check probe_scraper tests ./*.py
flake8 --max-line-length 100 probe_scraper tests ./*.py
yamllint repositories.yaml .circleci
python -m isort --profile black --check-only probe_scraper tests ./*.py

To render API documentation locally to index.html:

make apidoc

Developing using the container

Run tests in container. This does not run tests that require a web connection:

export COMMAND='pytest tests/'
make run

To run all tests, including those that require a web connection:

make test

To test whether the code conforms to the style rules, you can run:

make lint

Tests with Web Dependencies

Any tests that require a web connection to run should be marked with @pytest.mark.web_dependency.

These will not run by default, but will run on CI.

Performing a Dry-Run

Before opening a PR, it's good to test the code you wrote on the production data. You can specify a specific Firefox version to run on by using first-version:

export COMMAND='python -m probe_scraper.runner --firefox-version 65 --dry-run'
make run

or locally via:

python -m probe_scraper.runner --firefox-version 65 --dry-run

Including --dry-run means emails will not be sent.

Additionally, you can test just on Glean repositories:

export COMMAND='python -m probe_scraper.runner --glean --dry-run'
make run

By default that will test against every Glean repository, which might take a while. If you want to test against just one (e.g. a new repository you're adding), you can use the --glean-repo argument to just test the repositories you care about:

export COMMAND='python -m probe_scraper.runner --glean --glean-repo glean-core --glean-repo glean-android --glean-repo burnham --dry-run'
make run

Replace burnham in the example above with your repository and its dependencies.

You can also do the dry-run locally:

python -m probe_scraper.runner --glean --glean-repo glean-core --glean-repo glean-android --glean-repo burnham --dry-run

Module overview

The module is built around the following data flow:

  • scrape registry files from mozilla-central, clone files from repositories directory
  • extract probe data from the files
  • transform probe data into output formats
  • save to disk

The code layout consists mainly of:

  • probe_scraper
    • runner.py - the central script, ties the other pieces together
    • scrapers
      • buildhub.py - pull build info from the BuildHub service
      • moz_central_scraper.py - loads probe registry files for multiple versions from mozilla-central
      • git_scraper.py - loads probe registry files from a git repository (no version or channel support yet, just per-commit)
    • parsers/ - extract probe data from the registry files
    • transform_*.py - transform the extracted raw data into output formats
  • tests/ - the unit tests

Accessing the data files

The processed probe data is serialized to the disk in a directory hierarchy starting from the provided output directory. The directory layout resembles a REST-friendly structure.

|-- product
    |-- general
    |-- revisions
    |-- channel (or "all")
        |-- ping type
            |-- probe type (or "all_probes")

For example, all the JSON probe data in the main ping for the Firefox Nightly channel can be accessed with the following path: firefox/nightly/main/all_probes. The probe data for all the channels (same product and ping) can be accessed instead using firefox/all/main/all_probes.

The root directory for the output generated from the scheduled job can be found at https://probeinfo.telemetry.mozilla.org/. All the probe data for Firefox coming from the main ping can be found at https://probeinfo.telemetry.mozilla.org/firefox/all/main/all_probes.

Accessing Glean metrics data

Glean data is generally laid out as follows:

| -- glean
    | -- repositories
    | -- general
    | -- repository-name
        | -- general
        | -- metrics

For example, the data for a repository called fenix would be found at /glean/fenix/metrics. The time the data was last updated for that project can be found at glean/fenix/general.

A list of available repositories is at /glean/repositories.